Development of Fault Detection and Diagnosis Model for Drilling Machines

Drilling machines are essential in industrial applications as they are used to drill materials such as metal, plastic, and concrete and are now being incorporated into smart industries. Such machinery needs to be maintained properly given that they are known to wear out very quickly. In its previous form, as a manual process, monitoring has served its purpose significantly. Nowadays, it is replaced by automated systems that utilize achievements in signal processing and machine learning. This work proposes fault detection for drilling machines through sound signals and Fine K Nearest Neighbour (Fine KNN). Fine KNN was selected due to its moderate accuracy and computational efficiency compared to other classifiers in real-time despite a slightly lower accuracy than Quadratic SVM or bagged trees. The dataset employed is obtained from a GitHub repository and contains sound signals under various fault scenarios: healthy, bearing, gear, and fan. In feature extraction, 16 time-domain and frequency-domain features are extracted from the chosen signal and then narrowed down to 12 by using the RelieF algorithm to improve the model. The Fine KNN model maintains an efficiency of operation while detecting faults at a rate of 95.6%, which indicates the model’s accuracy. For this reason, feature selection and preprocessing serve a critical role in enhancing the model performance and suitability for real-time applications as affirmed by this research. Thus, this research opens up the possibility of integrating more complex models for machine condition monitoring at the edge devices. Future work will focus on obtaining more sophisticated classifiers and better preprocessing for improving fault detection performances in compact and power-efficient platforms suitable for the industrial IoT environment.